代码审查
- 作者仓库星标 175
- 作者仓库 pai-opencode
- 领域
- 安全
- 作者声明的智能体
- 未找到明确声明;不据此推断已兼容或已测试
- 静态检查
- 88 / 100 · 启发式扫描,不代表运行安全
- 运行环境证据
- 未执行;当前没有隔离运行测试证据
- 作者 / 版本 / 许可证
- @Steffen025 · 未声明许可证
- 上下文用量评级
- 中等消耗
- 接入复杂程度
- 需简单配置
- 是否需要外部接口密钥
- 不需要
- 兼容的系统
- macOS · Linux
- 运行环境要求
- 未检出明确要求
- 检测到的文件与系统行为
-
- 只读
- 命令执行
- 网络行为
- 仅限本地
- 安装命令数
- 无(仅作为资料)
需要注意: 未声明工具权限范围;实际权限由运行环境决定。
# CodeReview Skill
Use this skill to run AI-powered code review via **roborev** — a local, MIT-licensed review
tool with explicit OpenCode support.
---
## USE WHEN
- Completing a BUILD phase and want to catch issues before committing
- Running VERIFY phase and need reproducible quality evidence
- Reviewing a PR diff before merging
- Auditing plugin handler code for pattern violations (no `console.log`, handler structure)
- You want AI-powered architectural review of your changes
- Trigger phrases: "review code", "check my changes", "roborev", "code quality check"
---
## MANDATORY
1. **roborev must be installed** — `brew install roborev-dev/tap/roborev`
2. **One-time init** — `roborev init` (installs post-commit git hook)
3. **`.roborev.toml` must exist** at repo root with `agent = "opencode"` and `review_guidelines`
4. **Invoke** via `code_review` tool (preferred) or `roborev review --dirty` in EXECUTE phase
5. **Cite exit code 0** as VERIFY evidence: `code_review tool returned exit 0 — no findings`
---
## OPTIONAL
- `roborev skills install` — installs the roborev OpenCode skill (adds roborev commands to agent)
- `mode: "last-commit"` — review the last git commit instead of dirty working tree
- `mode: "fix"` — feed findings to the agent for automatic fixes
- `mode: "refine"` — run auto-fix loop until review passes
- `path` argument — focus review on a specific file or glob
---
## What roborev Does
roborev analyzes your staged/uncommitted changes (or last commit) using an LLM and surfaces:
- Code quality issues
- Security concerns
- Architectural violations
- Style inconsistencies
- Bugs and edge cases
All review runs locally. No account or cloud service required.
---
## Quick Reference
```bash
# Review uncommitted changes (most common)
roborev review --dirty
# Review last commit
roborev review
# Feed findings to agent for fixes
roborev fix
# Auto-fix loop until review passes
roborev refine
# Install git post-commit hook (one-time)
roborev init
# Install OpenCode skill for roborev
roborev skills install
```
---
## Algorithm Integration
### When to invoke
The Algorithm invokes code review in two ways:
**1. Via `code_review` tool (plugin-provided)**
Call the `code_review` tool directly from any Algorithm phase:
```text
Use code_review tool with mode="dirty" to review uncommitted changes before commit.
```
**2. Via `roborev` CLI in EXECUTE/VERIFY phase**
```bash
# In EXECUTE: review before committing
roborev review --dirty
# In VERIFY: evidence that review passed
roborev review --dirty && echo "PASS" || echo "FINDINGS"
```
### Recommended Algorithm workflow
```text
BUILD → commit changes
EXECUTE: roborev review --dirty
→ If PASS: continue to VERIFY
→ If FINDINGS: address in next BUILD iteration
VERIFY: cite roborev exit code 0 as evidence
```
---
## Installation
### macOS / Linux (Homebrew)
```bash
brew install roborev-dev/tap/roborev
```
### Go
```bash
go install github.com/roborev-dev/roborev@latest
```
### Verify
```bash
roborev --version
```
---
## One-Time Setup
```bash
# 1. Install git post-commit hook (auto-reviews on every commit)
roborev init
# 2. Install OpenCode skill (adds roborev commands to agent)
roborev skills install
# 3. Verify config exists at repo root
cat .roborev.toml
```
---
## Configuration (`.roborev.toml`)
This repo's config is at `.roborev.toml` in the root. Key settings:
```toml
agent = "opencode"
review_guidelines = """
# PAI-OpenCode Review Guidelines
...
"""
```
The `review_guidelines` field gives roborev domain-specific rules for this project —
including the no-console.log constraint, file-logger pattern, and model routing rules.
---
## Troubleshooting
### roborev not found
```bash
# Install via Homebrew
brew install roborev-dev/tap/roborev
# Or check if it's in PATH
which roborev
echo $PATH
```
### Review times out
Default timeout is 2 minutes. For large changesets, focus the review:
```bash
roborev review --dirty -- src/specific/file.ts
```
### No changes found
Make sure you have uncommitted changes:
```bash
git diff
git diff --cached # staged changes
```
### Post-commit hook not running
Re-run `roborev init` to reinstall the hook:
```bash
cat .git/hooks/post-commit # verify hook exists
roborev init # reinstall if missing
```
---
## PAI-OpenCode Specific Guidelines
When running code review on this project, roborev checks for:
1. **No console.log** — all plugin logging via `fileLog()` / `fileLogError()`
2. **Handler pattern** — new capabilities = new handler file + import + registration
3. **No hardcoded models** — model routing via `opencode.json` only
4. **TypeScript strict** — no implicit any, explicit return types on exports
5. **Biome formatting** — tabs, 100 char line width, double quotes
---
## Related
- `.roborev.toml` — project review configuration
- `ADR-018` — architectural decision for roborev integration
- `.opencode/plugins/handlers/roborev-trigger.ts` — plugin handler providing `code_review` tool
- `.github/workflows/code-quality.yml` — CI pipeline (Biome check on PRs)
章节与文件索引
按说明文件中的顺序排列。
使用时间
强制
可选
罗博列夫做什么
何时调用
文件/命令 -> console.log · brew install roborev-dev/tap/roborev · roborev init · .roborev.toml · agent = "opencode" · review_guidelines · code_review · roborev review --dirty
内容校验值 -> a184d0712ec1
当前原文的自动摘译,尚未人工审校。
使用时间
完成构建阶段并希望在提交运行验证阶段之前捕获问题并需要可重现的质量证据在合并之前检查 PR 差异审核插件处理程序代码是否存在模式违规(无
console.log,处理程序结构)您希望对更改进行 AI 驱动的架构审查强制
roborev 必须安装 —
brew install roborev-dev/tap/roborev一次性初始化 —roborev init(安装提交后 git hook).roborev.toml必须与agent = "opencode"和review_guidelines一起存在于仓库根目录 通过code_review工具(首选)或roborev review --dirty在 EXECUTE 阶段调用 引用退出代码 0作为验证证据:code_review tool returned exit 0 — no findings可选
roborev skills install— 安装 roborev OpenCode 技能(向代理添加 roborev 命令)mode: "last-commit"— 检查最后一次 git 提交,而不是脏工作树mode: "fix"— 将结果反馈给代理以进行自动修复mode: "refine"— 运行自动修复循环,直到审核通过path参数 — 重点审核特定文件或 glob罗博列夫做什么
roborev 使用 LLM 分析您的分阶段/未提交的更改(或上次提交)并显示: 代码质量问题 安全问题 架构违规 风格不一致
何时调用
该算法通过两种方式调用代码审查: 1. 通过
code_review工具(插件提供) 直接从任何算法阶段调用code_review工具: 2. 在 EXECUTE/VERIFY 阶段通过roborevCLI